CompoDiff: Versatile Composed Image Retrieval With Latent Diffusion

Benchmark Model Rank Results
zero-shot-composed-image-retrieval-zs-cir-on-circoCompoDiff (CLIP G/14)#26mAP@10: 17.71
zero-shot-composed-image-retrieval-zs-cir-on-circoCompoDiff (CLIP L/14)#31mAP@10: 13.51
zero-shot-composed-image-retrieval-zs-cir-on-cirrCompoDiff (CLIP G/14)#27R@5: 57.61
zero-shot-composed-image-retrieval-zs-cir-on-cirrCompoDiff (CLIP L/14)#37R@5: 54.36
zero-shot-composed-image-retrieval-zs-cir-on-fashion-iqCompoDiff (CLIP G/14)#10(Recall@10+Recall@50)/2: 45.37
zero-shot-composed-image-retrieval-zs-cir-on-fashion-iqCompoDiff (CLIP L/14)#15(Recall@10+Recall@50)/2: 44.11